s390/vmcp: make use of contiguous memory allocator

If memory is fragmented it is unlikely that large order memory
allocations succeed. This has been an issue with the vmcp device
driver since a long time, since it requires large physical contiguous
memory ares for large responses.

To hopefully resolve this issue make use of the contiguous memory
allocator (cma). This patch adds a vmcp specific vmcp cma area with a
default size of 4MB. The size can be changed either via the
VMCP_CMA_SIZE config option at compile time or with the "vmcp_cma"
kernel parameter (e.g. "vmcp_cma=16m").

For any vmcp response buffers larger than 16k memory from the cma area
will be allocated. If such an allocation fails, there is a fallback to
the buddy allocator.

Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>

6 files changed, 90 insertions, 9 deletions

diff --git a/Documentation/admin-guide/kernel-parameters.txt b/Documentation/admin-guide/kernel-parameters.txt
index d9c171ce4190..5a2d5079139b 100644
--- a/Documentation/admin-guide/kernel-parameters.txt
+++ b/Documentation/admin-guide/kernel-parameters.txt
@@ -4375,6 +4375,10 @@
                         decrease the size and leave more room for directly
                         mapped kernel RAM.
 
+        vmcp_cma=nn[MG] [KNL,S390]
+                        Sets the memory size reserved for contiguous memory
+                        allocations for the vmcp device driver.
+
         vmhalt=         [KNL,S390] Perform z/VM CP command after system halt.
                         Format: <command>
 
diff --git a/arch/s390/include/asm/setup.h b/arch/s390/include/asm/setup.h
index 61da4bd6edad..490e035b3716 100644
--- a/arch/s390/include/asm/setup.h
+++ b/arch/s390/include/asm/setup.h
@@ -108,6 +108,12 @@ extern void pfault_fini(void);
 #define pfault_fini()           do { } while (0)
 #endif /* CONFIG_PFAULT */
 
+#ifdef CONFIG_VMCP
+void vmcp_cma_reserve(void);
+#else
+static inline void vmcp_cma_reserve(void) { }
+#endif
+
 void report_user_fault(struct pt_regs *regs, long signr, int is_mm_fault);
 
 void cmma_init(void);
diff --git a/arch/s390/kernel/setup.c b/arch/s390/kernel/setup.c
index a50238e17867..164a1e16b53e 100644
--- a/arch/s390/kernel/setup.c
+++ b/arch/s390/kernel/setup.c
@@ -925,6 +925,7 @@ void __init setup_arch(char **cmdline_p)
         setup_memory_end();
         setup_memory();
         dma_contiguous_reserve(memory_end);
+        vmcp_cma_reserve();
 
         check_initrd();
         reserve_crashkernel();
diff --git a/drivers/s390/char/Kconfig b/drivers/s390/char/Kconfig
index b3f1c458905f..97c4c9fdd53d 100644
--- a/drivers/s390/char/Kconfig
+++ b/drivers/s390/char/Kconfig
@@ -169,10 +169,21 @@ config VMCP
         def_bool y
         prompt "Support for the z/VM CP interface"
         depends on S390
+        select CMA
         help
           Select this option if you want to be able to interact with the control
           program on z/VM
 
+config VMCP_CMA_SIZE
+        int "Memory in MiB reserved for z/VM CP interface"
+        default "4"
+        depends on VMCP
+        help
+          Specify the default amount of memory in MiB reserved for the z/VM CP
+          interface. If needed this memory is used for large contiguous memory
+          allocations. The default can be changed with the kernel command line
+          parameter "vmcp_cma".
+
 config MONREADER
         def_tristate m
         prompt "API for reading z/VM monitor service records"
diff --git a/drivers/s390/char/vmcp.c b/drivers/s390/char/vmcp.c
index b5e3a49745f9..c202b407698f 100644
--- a/drivers/s390/char/vmcp.c
+++ b/drivers/s390/char/vmcp.c
@@ -17,15 +17,77 @@
 #include <linux/kernel.h>
 #include <linux/miscdevice.h>
 #include <linux/slab.h>
+#include <linux/uaccess.h>
 #include <linux/export.h>
+#include <linux/mutex.h>
+#include <linux/cma.h>
+#include <linux/mm.h>
 #include <asm/compat.h>
 #include <asm/cpcmd.h>
 #include <asm/debug.h>
-#include <linux/uaccess.h>
 #include "vmcp.h"
 
 static debug_info_t *vmcp_debug;
 
+static unsigned long vmcp_cma_size __initdata = CONFIG_VMCP_CMA_SIZE * 1024 * 1024;
+static struct cma *vmcp_cma;
+
+static int __init early_parse_vmcp_cma(char *p)
+{
+        vmcp_cma_size = ALIGN(memparse(p, NULL), PAGE_SIZE);
+        return 0;
+}
+early_param("vmcp_cma", early_parse_vmcp_cma);
+
+void __init vmcp_cma_reserve(void)
+{
+        if (!MACHINE_IS_VM)
+                return;
+        cma_declare_contiguous(0, vmcp_cma_size, 0, 0, 0, false, "vmcp", &vmcp_cma);
+}
+
+static void vmcp_response_alloc(struct vmcp_session *session)
+{
+        struct page *page = NULL;
+        int nr_pages, order;
+
+        order = get_order(session->bufsize);
+        nr_pages = ALIGN(session->bufsize, PAGE_SIZE) >> PAGE_SHIFT;
+        /*
+         * For anything below order 3 allocations rely on the buddy
+         * allocator. If such low-order allocations can't be handled
+         * anymore the system won't work anyway.
+         */
+        if (order > 2)
+                page = cma_alloc(vmcp_cma, nr_pages, 0, GFP_KERNEL);
+        if (page) {
+                session->response = (char *)page_to_phys(page);
+                session->cma_alloc = 1;
+                return;
+        }
+        session->response = (char *)__get_free_pages(GFP_KERNEL | __GFP_RETRY_MAYFAIL, order);
+}
+
+static void vmcp_response_free(struct vmcp_session *session)
+{
+        int nr_pages, order;
+        struct page *page;
+
+        if (!session->response)
+                return;
+        order = get_order(session->bufsize);
+        nr_pages = ALIGN(session->bufsize, PAGE_SIZE) >> PAGE_SHIFT;
+        if (session->cma_alloc) {
+                page = phys_to_page((unsigned long)session->response);
+                cma_release(vmcp_cma, page, nr_pages);
+                session->cma_alloc = 0;
+                goto out;
+        }
+        free_pages((unsigned long)session->response, order);
+out:
+        session->response = NULL;
+}
+
 static int vmcp_open(struct inode *inode, struct file *file)
 {
         struct vmcp_session *session;
@@ -51,7 +113,7 @@ static int vmcp_release(struct inode *inode, struct file *file)
 
         session = file->private_data;
         file->private_data = NULL;
-        free_pages((unsigned long)session->response, get_order(session->bufsize));
+        vmcp_response_free(session);
         kfree(session);
         return 0;
 }
@@ -97,9 +159,7 @@ vmcp_write(struct file *file, const char __user *buff, size_t count,
                 return -ERESTARTSYS;
         }
         if (!session->response)
-                session->response = (char *)__get_free_pages(GFP_KERNEL
-                                                | __GFP_RETRY_MAYFAIL,
-                                                get_order(session->bufsize));
+                vmcp_response_alloc(session);
         if (!session->response) {
                 mutex_unlock(&session->mutex);
                 kfree(cmd);
@@ -146,9 +206,7 @@ static long vmcp_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
                 mutex_unlock(&session->mutex);
                 return put_user(temp, argp);
         case VMCP_SETBUF:
-                free_pages((unsigned long)session->response,
-                                get_order(session->bufsize));
-                session->response=NULL;
+                vmcp_response_free(session);
                 temp = get_user(session->bufsize, argp);
                 if (temp)
                         session->bufsize = PAGE_SIZE;
diff --git a/drivers/s390/char/vmcp.h b/drivers/s390/char/vmcp.h
index 1e29b0418382..4e725edf449f 100644
--- a/drivers/s390/char/vmcp.h
+++ b/drivers/s390/char/vmcp.h
@@ -20,8 +20,9 @@
 #define VMCP_GETSIZE _IOR(0x10, 3, int)
 
 struct vmcp_session {
-        unsigned int bufsize;
         char *response;
+        unsigned int bufsize;
+        unsigned int cma_alloc : 1;
         int resp_size;
         int resp_code;
         /* As we use copy_from/to_user, which might     *